Gpm motion refinement

Abstract

It provides a solution for video processing. A method for video processing is proposed. The method comprises: determining motion information for a video block of a video by conducting a refinement process on at least one of a first merge candidate or a second merge candidate for the video block, the second merge candidate being generated based on the first merge candidate; and performing a conversion between the video block and a bitstream of the video based on the motion information. Compared with the conventional solution, the proposed method can advantageously improve the coding performance.

1 - 41 . (canceled) 
     
     
         42 . A method for video processing, comprising:
 determining motion information for a video block of a video by conducting a refinement process on at least one of a first merge candidate or a second merge candidate for the video block, the second merge candidate being generated based on the first merge candidate; and   performing a conversion between the video block and a bitstream of the video based on the motion information.   
     
     
         43 . The method of  claim 42 , wherein determining motion information for a video block comprises:
 constructing a merge candidate list for the video block, the merge candidate list comprising at least the second merge candidate generated based on the refinement process conducted on the first merge candidate; and   determining the motion information based on the merge candidate list.   
     
     
         44 . The method of  claim 43 , wherein determining the motion information based on the motion information comprises: conducting a further refinement process on the second merge candidate to determine the motion information; or
 wherein the refinement process comprises refining at least one of: L0 motion information of the first merge candidate; and L1 motion information of the first merge candidate; or   wherein the first merge candidate comprises a bi-prediction merge candidate or a uni-prediction merge candidate; or   wherein the refinement process is enabled if the first merge candidate satisfies a condition associated with a corresponding refinement method.   
     
     
         45 . The method of  claim 42 , wherein determining motion information for a video block comprises:
 constructing a merge candidate list for the video block, the merge candidate list comprising at least the second merge candidate generated based on the first merge candidate; and   conducting the refinement process on the second merge candidate to determine the motion information for the video block.   
     
     
         46 . The method of  claim 42 , wherein the video block comprises at least two parts, and wherein the refinement process comprises motion refinement processes simultaneously conducted for the at least two parts; or
 wherein the video block comprises at least two parts, and wherein the refinement process comprises motion refinement processes conducted for the at least two parts respectively; or   wherein the video block comprises at least two parts, and wherein the refinement process is applied to one or more parts of the at least two parts.   
     
     
         47 . The method of  claim 46 , wherein the at least one or more parts are predefined or selected according to a rule. 
     
     
         48 . The method of  claim 42 , wherein the refinement process is based on bilateral matching. 
     
     
         49 . The method of  claim 48 , wherein the video block comprises a geometric partition mode (GPM) block, and L0 and/or L1 prediction in the bilateral matching of the GPM block is irrelevant to a GPM split mode of the GPM block; or
 wherein the video block comprises a GPM block, and the L0 and/or L1 prediction in the bilateral matching of the GPM block is based on a GPM split mode of the GPM block.   
     
     
         50 . The method of  claim 42 , wherein the refinement process is based on based on template matching. 
     
     
         51 . The method of  claim 50 , wherein a template used in the template matching is uni-directional or bi-directional; or
 wherein the video block comprises at least two parts, and templates for different parts are determined based on different rules; or   wherein the template matching is applied to the whole video block for determining partial motion information associated with one part of the video block; or   wherein the video block comprises at least two parts, and the template matching is applied to a target part of the at least two parts.   
     
     
         52 . The method of  claim 51 , wherein a shape of a template for the target part is determined based on a shape of the target part. 
     
     
         53 . The method of  claim 42 , wherein whether the refinement process is based on bilateral matching or template matching is determined based on first motion information of the first merge candidate; or
 wherein the refinement process is enabled if the video block is coded in GPM mode; or   wherein the bitstream comprises a flag indicating whether the refinement process is enabled;   wherein at least one of the first or second merge candidate is used for subblock based motion vector storage for the video block; or   wherein the first or second motion candidate is used for deblocking strength determination for the video block; or   wherein the method further comprises: performing motion compensation for the video block based on the first merge candidate or the second merge candidate; or   wherein the video block comprises a first video block, and the method further comprises:   generating an advanced motion vector predication (AMVP) candidate list or a merge candidate list for a second video block of the video based on the motion information.   
     
     
         54 . The method of  claim 53 , wherein the motion information is further used as:
 a temporal motion vector candidate if the first video block is a temporal neighbor block of the second video block, or   a spatial motion vector candidate if the first video block is a spatial neighbor block of the second video block.   
     
     
         55 . The method of  claim 42 , wherein the video block comprises a first video block, and the method further comprises:
 generating an AMVP candidate list or a merge candidate list for a second video block of the video block based on the first merge candidate for the first video block.   
     
     
         56 . The method of  claim 42 , wherein the video block is coded in geometric prediction mode with motion vector differences (GMVD) mode, and the method further comprises:
 adding a motion vector difference (MVD) to the motion information.   
     
     
         57 . The method of  claim 42 , wherein the video block is coded in GMVD mode and the motion information comprises first motion information, and wherein determining motion information for a video block comprises:
 adding a MVD to second motion information of the first merge candidate to derive third motion information, and   determining the first motion information by refining the third motion information.   
     
     
         58 . The method of  claim 42 , wherein the motion information is associated with a target coding mode, and the target coding mode comprises at least one of:
 geometric merge mode (GEO),   geometric partition mode (GPM),   wedge prediction mode,   triangular prediction mode (TPM),   geometric prediction mode with motion vector differences (GMVD), or   a GPM block with motion refinement; or   wherein the refinement process comprises at least one of:   a decoder side motion vector refinement (DVMR) process,   a frame-rate up conversion (FRUC) refinement process,   a template matching (TM) refinement process,   a merge mode with motion vector differences (MMVD) refinement process, or   a bi-directional optical flow (BDOF) refinement process.   
     
     
         59 . The method of  claim 42 , wherein the conversion comprises:
 decoding the video block from the bitstream; or encoding the video block into the bitstream.   
     
     
         60 . An apparatus for processing video data comprising a processor and a non-transitory memory with instructions thereon, wherein the instructions upon execution by the processor, cause the processor to:
 determine motion information for a video block of a video by conducting a refinement process on at least one of a first merge candidate or a second merge candidate for the video block, the second merge candidate being generated based on the first merge candidate; and   perform a conversion between the video block and a bitstream of the video based on the motion information.   
     
     
         60 . A non-transitory computer-readable storage medium storing instructions that cause a processor to:
 determine motion information for a video block of a video by conducting a refinement process on at least one of a first merge candidate or a second merge candidate for the video block, the second merge candidate being generated based on the first merge candidate; and   perform a conversion between the video block and a bitstream of the video based on the motion information.